Model
Model explorer

Poro 34B

OPEN
Silo AI (LumiOpen) · Poro family · released Apr 1, 2024

Fully open Apache-2.0 Finnish/English/code model trained on 1T tokens on the LUMI supercomputer; pivotal European low-resource open LLM.

ReasoningCodingVisionFunction callingTool useAgentic
57.8
Elo · rank #395
Parameters
34.2B
Active params
34.2B (dense)
Context
2K tokens
Architecture
Dense transformer, BLOOM architecture + ALiBi (54 layers, d_model 7168, 56 heads)
License
Apache-2.0
Languages
2+
API price (in/out)
No hosted API
Modalities
text
Benchmark results
Bar shows position within the tracked field; marker = field best
ARC-ChallengeReasoning53.2%#113
best: Llama 3.1 405B · 96.9%
FIN-benchKnowledge66.3%#1
best: this model · 66.3%
GSM8KMath11.8%#175
best: Llama 3.1 405B · 96.8%
HellaSwagReasoning77.8%#92
best: Claude 3 Opus · 95.4%
HumanEvalCoding37.2%#139
best: Claude Opus 4.5 · 99.4%
MBPPCoding47.4%#75
best: Llama-3.3-Nemotron-Super-49B v1 (Reasoning On) · 91.3%
MMLUKnowledge46.3%#241
best: OpenAI o3 · 92.9%
TruthfulQAKnowledge41.7%#74
best: Phi-3.5-MoE (16x3.8B, 6.6B active) · 77.5%
WinoGrandeReasoning72.8%#82
best: PaLM 2 · 90.9%
Run it locally
VRAM @ Q4
20 GB
VRAM @ FP16
68 GB
Fits on (Q4)
RTX 3090 24GBRTX 4090 24GBRTX 5090 32GBM4 Pro 48GBM3 Max 128GBM3 Ultra 512GBA100 80GBH100 80GBH200 141GBB200 192GB
Q4 GGUF (~20GB) fits on a single 24GB RTX 4090 with limited headroom for context.
Quantizations
GGUF · AWQ
Poro family
Elo progression across releases
Poro 34BApr 202457.8
Resources
HFLumiOpen/Poro-34BDOCSarxiv.org/abs/2404.01856
API price weights · each benchmark row carries its own source badge (see methodology)